01 Light matters more than the tool

Nine out of ten AI-generated images look like cheap plastic. The cause is almost never the AI you chose: it’s the light no one described. In cinema, there’s a rule of thumb — lighting accounts for about 70% of what makes a scene look expensive. The tool accounts for the other 30%.

The photographer knows this by heart: the same face, photographed at noon in a parking lot and near a window in the late afternoon, looks like two different people. The camera is the same. The light is what changed.

It’s the same with AI. When you write a prompt—the so-called prompt — without mentioning lighting, the AI chooses for you. And it picks the dullest lighting possible.

02 Too perfect looks fake

Here’s the paradox that separates amateurs from professionals: skin with no pores, fabric with no folds, a floor with no marks — all of it screams “computer.” The human eye trusts imperfection because the real world is imperfect.

That’s why people who work seriously with images do the opposite of what instinct suggests: they ask for texture, grain, visible pores, and small signs of wear. They deliberately make the image “worse”—and it starts to look like an expensive photograph.

A social media manager sees this in practice: a post that’s too perfect looks like a bank ad, and the audience scrolls right past. A photo with a real-life texture makes them stop.

03 The professional formula: observe, describe, reference

Every cinematic result comes from the same three-part recipe: observation (you look at a photo you admire and notice the lighting, lens, and mood), technical description (you turn what you saw into precise words in the prompt) and reference (you provide your own photo for the AI to copy the color palette and style).

A reference image is the most powerful shortcut of the three: instead of trying to explain a color tone with words, you show it.

A restaurant owner can use this tomorrow: take the cover photo of an award-winning restaurant he admires, provide it as a reference, and ask for his dish in that same window light with soft shadows.

04 Let your chat AI be your cinematographer

"But I don't know how to talk about lenses and shadows" — you don't need to. You'll use a chat AI (ChatGPT works, even in the free version) as a translator: it looks at the reference photo and writes the technical description for you.

The process is simple: you send the photo you admire to ChatGPT along with a ready-made request, and it returns two things — an analysis of the photo in a photographer's language and a technical prompt ready to paste into any image generator.

For the photographer, it’s like having an assistant who can describe any light she shows them. For the social media manager, it means turning any client reference into a technical request in a minute.

The ready-to-use prompt is in the exercise just below. It’s a recipe: you don’t need to understand every word—just swap in the photo.

01 AI has no memory from one prompt to the next

Every time you ask for a new image, the AI starts from scratch — it doesn’t remember the face it created a minute ago. Asking for “the same woman, now in profile” is like asking an artist who’s never seen her to draw her. That’s why the face changes.

The professionals’ solution is clever: instead of six separate requests, a single request that already brings all six angles together, on the same sheet. Since everything is created in the same generation, the AI keeps the face, clothes, and lighting consistent across all the frames.

For the photographer, it’s the equivalent of a whole shoot on a single contact sheet: instead of six sessions with six similar models, one contact sheet with the same model in six poses.

02 The six-angle grid, in a ready-to-use request

The prompt below is a tested recipe: upload your image from lesson 1, paste the text, and get back a sheet with six photos of the same scene—close-up, medium shot, and wide shot—as if a photographer had walked around the set.

It’s in English because image generators understand the technical terms better that way. You don’t need to translate it or understand it word for word — the entire recipe applies as is.

The restaurant owner can use this tomorrow: one good photo of the main dish becomes six angles of the same dish—from above for the menu, from the side for the post, and a close-up of the details for the ad.

03 Why the sheet is square

It may seem like a detail, but it’s geometry: every vertical film frame is taller than it is wide. To fit six of those frames side by side on a single sheet, the sheet has to be square—three columns, two rows, everything fitting snugly.

If you leave the aspect ratio vertical, the AI tries to squeeze the grid into a narrow corridor: cropped frames, repeated angles, an unusable sheet.

And there’s a second key in the prompt: it prohibits AI from reinventing. No changing clothes, changing hairstyles, or "improving" the setting. Social media managers know this pain: the client approved a look, and the entire series of posts needs to keep that exact look.

04 The method has limits—and it’s good to know that now

Honesty: in the wider frames of the grid, the face looks small and loses detail. Professional methods for total consistency exist—you’ll see them later in the program. The grid is the method fast and hassle-free to get started today.

Think of it as the working base: the frames that become video in the next lesson come from it. For the photographer, it’s the shoot’s draft, not the final photo delivered to the client — and a good draft is what separates a productive shoot from a wasted afternoon.

01 Stop guessing: lock the beginning and the end

The secret of this lesson fits in one sentence: instead of giving an image and hoping for the best, you provide two — the frame where the clip begins and the frame where it ends. The AI only fills in the transition between them.

With both endpoints locked in, the identity has nowhere to drift: the face at the beginning and the face at the end are the same, so the middle stays on track.

The social media manager knows the manual version of this: in a client video, she sets the first and last scenes before anything else—the middle exists to connect them. It’s the same here, except the AI creates the middle.

02 Hand-crop the two frames

Using your six-angle sheet (lesson 2), you’ll crop two frames—for example, the medium shot and the close-up—each in a vertical phone format (9:16 aspect ratio, the same as stories and reels).

Hand-cropping before handing it over to AI is a directing choice: you choose the framing instead of letting the machine cut wherever it wants. The photo editor built into your phone will do the job.

The photographer has always done this: on the contact sheet, she marks the two frames that work with a pen—no one gives the client the whole sheet.

03 In the video prompt, describe the camera — not the person

Leaving the prompt blank is the mistake that makes the video “melt”: without instructions, AI invents movement for everything—face, clothes, background. The professional formula reverses the logic: camera movement + minimal subject action + atmosphere.

A ready-made example, in the spirit of what you’ll paste:

Cinematic slow push-in, smooth zoom on the face,
the character remains stationary, warm lighting.
No morphing, consistent features.

Breaking down the recipe: the camera moves forward slowly, the person barely moves, the light stays warm—and the last two words forbid AI from distorting the face.

Don’t know camera terms? Ask the chat AI: "write 3 camera movement prompts in English to connect a medium shot to a close-up in 5 seconds". A restaurant owner doesn’t need to study filmmaking — they need to know who to ask.

04 The chain: clips that join together on their own

A 5-second clip is too short to tell a story. The solution is called a chain: the last frame from clip A becomes the first frame from clip B. Since the splice happens on an identical frame, the cut is invisible — and the sequence can grow without limit.

For the restaurant owner, it’s one continuous tour: clip 1 moves through the dining room and ends at the kitchen door; clip 2 starts at that exact door and reveals the lit stove.

01 Pacing is math, not talent

If every segment of your video moves at the same speed, the audience falls asleep—it’s physiology, not opinion. The human eye wakes up with contrast: slow to show, fast to cut, slow again to breathe.

The structure you’ll use is a three-part formula: slow entry → acceleration during the cut → slow resolution. Every cut between clips happens inside the fast section, where the eye can’t inspect it.

Social media has seen this countless times without putting a name to it: the videos that hold attention to the end are the ones that change pace; the ones that lose viewers in three seconds race along flatly, at one speed.

02 The speed curve in practice

In CapCut—the free editor for your phone or computer—this formula has a button name: speed curve. You build the variation into the clip itself: start at 1x, speed up to 5x or 10x near the cut, and the next clip drops back to normal.

The photographer uses exactly this when making a behind-the-scenes video: the scene’s setup runs in fast motion, then the final click plays at normal speed—the audience feels the climax without knowing why.

03 Sound is half the illusion

The image catches the eye; sound holds attention. A silent video, no matter how beautiful, feels unfinished. The minimum recipe has two layers: one background track e two sound effects marking the transitions — a breath during the acceleration, a beat at the cut.

And the golden rule of mixing: when the effect hits, the music lower. Both things at full volume become noise; the effect over muffled music becomes impact.

Don’t know what effect to use? Describe the video to the chat AI and ask: "suggest 5 sound effects and where to place each one on the timeline". The restaurant owner did this for the dish video — they even got a tip to add the sizzle of the frying pan at the opening.

04 The right screen from the very first click

Before dragging in any clips, define the project format: 9:16, the vertical phone portrait—the same one as your clips from previous lessons. In CapCut, this is under Aspect Ratio (or Format) when you create the project.

For export, the suggested default is fine: 1080p resolution and everything else as is. Export, watch it on your phone, and ask yourself: can you see the seams between the clips? If not, the lesson delivered on its promise.

The social media manager learned this order the hard way: a finished edit in the wrong format means reframing every shot.

01 Generic prompt, generic face

Write “beautiful woman, realistic photo” and the AI gives you exactly what you asked for: the average of all the beautiful faces it knows. Average has no story. The problem isn’t aesthetic—it’s narrative: no one cares what happens to a stock-photo face.

The social media manager sees it in the numbers: a post with an “AI-perfect” face gets scrolled past; a post with a character who seems alive—a laugh line, an outfit with a story—makes people stop scrolling.

02 The nine layers, in four questions

Professional construction stacks nine layers. It seems like a lot, but they answer just four questions you already know how to ask:

• Who is it? — identity (age, features, presence), realistic skin texture (pores, natural shine, small marks), and wardrobe (materials, colors, accessories).
• What do you feel? — the emotional energy: calm, confident, mysterious, vulnerable. Emotion changes posture, gaze, and even what kind of light fits.
• Where is it? — the surroundings. This is the layer that turns a portrait into a movie scene.
• How is it filmed? — cinematic lighting (late afternoon, backlighting, shadows), camera language (lens, depth, composition), and the vertical 9:16 format.

The photographer recognizes her own process: it’s the mental sequence of a shoot—model, styling, location, light, lens. Only the medium is new: here, everything becomes text.

03 The ninth layer: say what you don’t want

The final layer is a list of prohibitions, written at the end of the prompt: no plastic skin, no computer-generated look, no distorted anatomy, no random objects, text, or watermarks in the image.

It works because when AI is unsure, it tends to fall back on exactly these bad habits—and an explicit prohibition closes the door before it can go in.

The restaurant owner creating his brand’s fictional “chef” can feel the difference right away: without the restrictions, you get a video game chef; with them, a professional who could be in his kitchen tomorrow.

04 The ready-to-use formula—and the protagonist who stays with us

The formula below organizes the nine layers into blanks. Fill in each bracket and paste it into ChatGPT’s image generator (the free version generates images)—or into the generator you already use.

For the next lessons, we created this the lighthouse keeper: a serene middle-aged woman on a stormy coast at dusk. She becomes a storyboard in lesson 6, motion in lesson 7, and a film in lesson 8. You can use her or create your own — there’s only one rule: choose someone you want to see come alive.

The social media manager can create a brand mascot for a client; the photographer can create a model who would be impossible to hire. The character is yours—the method is the same.

01 A beautiful image still isn’t a video

Between the portrait from lesson 5 and a clip that moves people is one thing beginners skip: deciding what happens, second by second. This technique has a film industry name— storyboard — but the idea is practical: it’s the shopping list for your scene.

And pay attention to the detail that changes everything: a storyboard isn’t a gallery of pretty variations. It’s a plan with timed beats, where each frame answers the question, "and now, what changes?"

The social media manager already does this without naming it when she sketches the sequence for a reel on paper: opening scene, turning point, ending. Here, the sketch becomes a professional nine-panel sheet.

02 The golden rule: everything stays still except the story

Across the nine frames, four things stay fixed: the character, the clothes, the world, and the light. Only one thing moves forward: the story. That’s the asymmetry that makes the sheet feel like scenes from the same film—and not nine different films.

The photographer recognizes the discipline behind a good shoot: same model, same outfit, same afternoon light throughout—the pose and intention are what evolve. When she changes the lighting midway, the client can tell “something broke,” even if they can’t say what.

03 The nine beats of 15 seconds

The arc you'll use is a condensed classic — it works for a storm at a lighthouse, a kitchen with the heat turned up, and a shop window:

• 0–1.5s · Opening silence — very wide shot, calm world.
• 1.5–3s · Something changes — wind, vapor, light: the atmosphere comes alive.
• 3–4.5s · The character feels — close-up on the face; the eyes notice.
• 4.5–6s · A detail moves — fabric, hand, ground texture.
• 6–7.5s · The tension builds — low angle, the world feels heavy.
• 7.5–9s · The hidden force — something appears in the distance.
• 9–10.5s · The mood intensifies — profile; the character turns around.
• 10.5–12.5s · The reveal — the main event takes place.
• 12.5–15s · Hero ending — low angle, character standing firm amid the chaos.

In the restaurant owner’s version: the kitchen is quiet, steam rises, the chef checks for doneness, flames lick the pan, and the finished dish arrives at the table like a revelation. Same arc, different world.

04 From the rule to the sheet: generating the storyboard

The generator builds the sheet for you: it takes your character image as a reference and the nine-beat arc as the script, then returns a 3×3 grid with a number, duration, title, and one action line per frame—with black borders like a professional contact sheet.

Your role is director: review the finished sheet frame by frame, checking the golden rule. Same face? Same clothes? Same world and lighting? Is the story moving forward?

The social media manager gets a sales asset here: the sheet approved by the client before before spending on any video generation — end of the conversation, "that wasn't what I imagined."

01 You direct; the assistant types.

A 15-second video with cinematic quality needs a huge prompt: each 1.5-second segment described in detail — what the character does, where the camera goes, how the light behaves. Writing all that by hand takes expert work.

The turning point: you don’t write the prompt—you commission the request. A chat AI, trained with the right instructions, reads your storyboard sheet and writes the entire technical page. You stay in charge of the decisions; it handles the paperwork.

Social media already works with this division of labor: they define the concept and references, then delegate the technical writing — except now the technical writer is free and responds in seconds.

02 The training: paste once, use forever

In a new ChatGPT conversation, you paste in a training text (it’s ready in the exercise). It turns that conversation into a video-prompt specialist with four habits that make all the difference:

• Read the timings on your sheet — if the storyboard says 0–1,5s, the prompt follows 0–1,5s.
• Detail each segment — action, camera, lighting, and physics, second by second.
• Protects the identity — repeat the instructions “same face, no distortion” in every prompt.
• Close with the technical tail — the anti-defect list (no melting, no flickering, realistic physics).

Save this conversation to your favorites: it’s a reusable tool you can use in every video project from now on. The photographer who puts together a portfolio animates a new scene every week—always in the same trained conversation.

03 Two references go in, one prompt comes out

With the assistant trained, you send the two pieces you created: the character image (lesson 5) and the storyboard sheet (lesson 6). Then you ask: “write the final prompt using both—following the timing on the sheet and preserving the identity.”

Together, the two complement each other: the portrait ensures a high-quality face; the sheet maps out what happens each second. One without the other falls short — the portrait alone has no story, and the sheet alone has faces too small to preserve identity.

The restaurant owner sends the fictional chef’s portrait and the kitchen sheet in nine beats—and gets back a technical page he would never write himself. He doesn’t need to: every decision was his.

04 Generate—and check against the sheet

The generation runs in a tool that supports Seedance 2.0—Higgsfield and CapCut are two options; if you already use another Seedance platform, the same request works there. You upload the two images, paste the request, adjust three options, and generate.

Next comes the director’s job: watch with the sheet beside you. The video isn’t good “because it looks beautiful”—it’s good if it followed the plan. That’s how the photographer checks a finished shoot: the question is never “does it look beautiful?” but “is this what the client approved on the sheet?”

01 Think like an editor, not a collector

The beginner’s mistake at this stage has a name: attachment. They spent generations producing each clip and want to use everything. The professional editor does the opposite—they watch the footage and ask just one thing: which moments serve the story? Everything else, no matter how beautiful, falls flat.

A case to get a feel for the method: Vera, a photographer, generated three scenes of the lighthouse keeper and hated the entire second clip — except for the two seconds when the cape billows in the wind. She cut the remaining thirteen seconds without a second thought. Those exact two seconds became the best transition in her film.

Structure for prospecting: preparation (the world and the threat appear) → conflict (the tension builds) → climax (the strongest moment, and the resolution). Every nugget you save will go in one of these three boxes.

02 Before editing, improve the raw material

AI-generated video loses fine detail — faces, fabric, dust — especially during fast movement. That’s why professionals run clips through a smart upscaling before editing.

The tool mentioned in this track is Topaz, a paid program with models that restore fabric and skin texture. A practical rule: if the result looks too sharp and jagged, switch to the tool’s own general cleanup model.

And the honest version for anyone just starting out: this step is optional. A social media manager who doesn’t want to pay for any tools yet can put the film together with the original clips—the lesson’s method works just as well; the upscaling just adds polish.

03 Break it into pieces, reassemble the story

In the editor, the professional workflow follows an order that saves hours: first, cut each clip into small pieces; then choose what to keep; only then put it together. Anyone who tries to “edit it all at once” is stuck with the clips in their original order.

The restaurant owner edits the film about his fictional chef by cutting the three clips into twelve pieces, throwing out seven—strange movements, empty seconds, broken frames—and reassembling five in the order that works: calm kitchen, rising flame, the exact moment, plating, the table.

04 Inherited sound, light polish, export

Good news about audio: today’s video generations already come with their own sound and ambience, often good ones. Listen to the original audio before changing anything — if it works, keep it. Adjust only what’s needed: lower loud peaks, mute bad sections, add an effect where the transition calls for one (the wind and beat from lesson 4 still apply).

For the look, polish is seasoning, not a renovation. A starting point that works: contrast +4, sharpness +10, saturation +4, highlights −7, temperature +2. If the clip has already gone through smart upscaling, cut the sharpness in half or set it to zero.

The photographer recognizes the principle of photo developing: a good adjustment is one nobody notices—the three scenes simply start to look as if they were filmed on the same day. Export in 1080p and watch on your phone, from beginning to end, without pausing.